A third of teachers now use artificial intelligence (AI) tools such as ChatGPT to help with their work, double the number who used it five months ago. As well as lesson planning, creating resources and writing reports, they are using AI to respond to parent emails, write UCAS references and job adverts. Polling from also shows it has been used to check if pupils have used AI for an assignment, and to show them the limitations of the tools in class. Chief analyst Professor Becky Allen told ResearchED on Saturday that many respondents talked about 鈥渉ow transformative鈥 large language models (LLMs) of AI had been 鈥渋n reducing their workloads and helping them make better plans and better resources鈥. 鈥淚 like to think of them as being like the best kids we teach in our class. They get the A*s, they get over 90 per cent. But they don鈥檛 get everything right. 鈥淔or some people this is the slam dunk as to why you can鈥檛 use large language models: they make errors. And they do. What I would say is that they don鈥檛 make errors at random.鈥 AI use doubles in five months A recent survey of more than 9,000 teachers and leaders found 34 per cent reported using the tools to 鈥渉elp with school work鈥. When Teacher Tapp asked the same question in April, just 17 per cent reported doing the same thing. Teachers and leaders in private schools were more likely (49 per cent) than those in state schools (33 per cent) to use AI. It was also more likely to be deployed in ‘outstanding’ schools. Two in five teachers in their twenties reported using AI, compared with 26 per cent of those aged 50-plus. Men (44 per cent) were also more likely than women (31 per cent) to use it. The tools were also far more popular among English and science teachers than maths. The development of more sophisticated AI has prompted a debate about its potential benefits to education, but also fears about its misuse. Earlier this year, exam boards warned that chatbots might pose 鈥渟ignificant risks鈥 if used during assessments. But Gillian Keegan, the education secretary, has said AI could 鈥渢ransform鈥 teachers鈥 day-to-day work, and pointed to 鈥渟ome interesting experiments around marking鈥. Give prompts for better marking Allen said simply asking AI tools to 鈥渕ark and give feedback and grades鈥 was a 鈥減retty bad thing to do鈥. 鈥淕ive it a detailed prompt about how it should behave, who the pupils are, the text being studied. If it鈥檚 English literature, upload the chapter you鈥檙e asking questions on so that it鈥檚 actually got some text that is the correct thing. Becky Allen 鈥淕ive it a rubric that details the specific attributes for which it should be giving marks. Give it some example papers that have been marked already. Give it some feedback after it鈥檚 marked some, and then use that to then revise how it marks in the future. 鈥淏efore you say 鈥榯his can鈥檛 be done鈥, try this first and see how far you manage to get.鈥 Others were more sceptical of LLMs. In a recent blog post, Daisy Christodoulou, the founder of No More Marking, warned the 鈥渆rror rate and error type of LLMs limits their educational applications鈥. Their unreliability also 鈥渕eans they are not well-equipped to assess pupils’ work鈥. 鈥淲hat about helping teachers with resource creation and lesson planning? Again, I don鈥檛 think LLMs can operate independently. Teachers will need to check and re-check their outputs.鈥 She acknowledged AI might save teachers time, “but it is not a magic silver bullet for workload problems鈥. Marking and feedback trial planned Ministers have commissioned a project to explore whether generative AI could be used for 鈥渕arking and feedback鈥. But the Department for Education has refused to say more about its plans. Asked what the government was doing to encourage the use of AI in schools, Nick Gibb, the schools minister, said the project would aim to 鈥渂etter understand and test鈥 its possible uses鈥. The project, based on responses to a recent call for evidence, would 鈥渆xplore, for example, whether it can be used for marking and feedback鈥. 鈥淭his project will help us to build a robust evidence base to inform future policy and to further explore the opportunities this technology presents. It will involve AI experts and educators.鈥 Ministers intend to publish the findings in the spring.